|
In This Section
Purpose
This tutorial will help you install and use MPI.NET, a .NET
library that enables the creation of high-performance parallel
applications that can be deployed on multi-threaded workstations and
Windows clusters. MPI.NET provides access to
the Message Passing
Interface (MPI) in C# and all of the other .NET languages. MPI
is a standard for message-passing programs that is widely
implemented and used for high-performance parallel programs that
execute on clusters and supercomputers.
By the end of this tutorial, you should be able to:
- Install MPI.NET and its prerequisites.
- Write parallel MPI applications for deployment on Windows
workstations and clusters using point-to-point and collective
communication.
- Execute parallel MPI applications locally and on a cluster.
Other Tutorials
This tutorial is written for online reading and uses C# for all of its examples. Please see the main documentation page for tutorials formatted for printing/offline reading and tutorials using other languages (e.g., Python).
MPI Programming Model
The MPI programming model is, as its name implies, is based
on message
passing. In a message-passing system, different
concurrently-executing processes communicate by sending messages
from one to another over a network. Unlike multi-threading, where
different threads share the same program state, each of the MPI
processes has its own, local program state that cannot be observed
or modified by any other process except in response to a
message. Therefore, the MPI processes themselves can be as
distributed as the network permits, with different processes running
on different machines or even different architectures.
Most MPI programs are written with
the Single Program,
Multiple Data (SPMD) parallel model, where each of the processes
is running the same program but working on a different part of the
data. SPMD processes will typically perform a significant amount of
computation on the data that is available locally (within that
process's local memory), communicating with the other processes in
the parallel program at the boundaries of the data. For example,
consider a simple program that computes the sum of all of the
elements in an array. The sequential program would loop through the
array summing all of the values to produce a result. In a SPMD
parallel program, the array would be broken up into several
different pieces (one per process), and each process would sum the
values in its local array (using the same code that the sequential
program would have used). Then, the processes in the parallel
program would communicate to combine their local sums into a global
sum for the array.
MPI supports the SPMD model by allowing the user to easily launch
the same program across many different machines (nodes) with a
single command. Initially, each of the processes are identical, with
one distinguishing characteristic: each process is assigned
a rank, which uniquely identifies that process. The ranks of
MPI processes are integer values from 0 to P-1, where P is the
number of processes launched as part of the MPI program. MPI
processes can query their rank, allowing different processes in the
MPI program to have different behavior, and exchange messages with
other processes in the same job via their ranks.
|
